A sequence-based method to predict the impact of regulatory variants using random forest

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Most disease-associated variants from genome-wide association studies (GWAS) are in noncoding regions.
• Characterizing the functional impact and risk of these noncoding variants remains challenging.
• Effective computational models are needed to predict variant risk and interpret genome function alterations.

Purpose of the Study:

• To develop a computational method for predicting the risk of genetic variants.
• To interpret how variants affect genome function based solely on DNA sequences.
• To identify genetic risk factors for complex traits and diseases.

Main Methods:

• Developed kmerForest, a random forest classifier using k-mer counts for predicting accessible chromatin regions from DNA sequences.
• Incorporated sequence conservation features to improve prediction performance.
• Assessed k-mer feature importance and characterized single nucleotide polymorphism (SNP) risk by analyzing feature importance changes.

Main Results:

• kmerForest outperformed existing methods in distinguishing accessible chromatin regions.
• The model successfully discriminated between pathogenic and normal SNPs.
• The method prioritized SNPs enriched for FOXA1 binding sites in breast cancer cell lines.

Conclusions:

• Presented a novel sequence-based method for interpreting functional genetic variants.
• The k-mer based score effectively measures SNP impact on genome function.
• This approach aids in identifying genetic risk factors for complex diseases.